Currently, if the user requests aio=native, but forgets to choose a cache mode that sets O_DIRECT, that request is silently ignored and raw falls back to aio=threads.
Deprecate that behaviour so we can make it an error in future qemu versions. Signed-off-by: Kevin Wolf <kw...@redhat.com> --- v2: - Use error_printf() instead of fprintf(stderr, ...) so that the warning will appear on the HMP monitor instead of stderr for drive_add. Markus suggested error_report(), which would have the advantage of displaying the option string and position in the config file. However, the result was a line that was so full that it wasn't very prominent as a warning any more. Making the warning very visible was the reason for using fprintf in v1. After thinking more about it, I came to the conclusion that for a deprecation warning like this, being visible is more important than finding the bad option quickly. The hope is anyway that nobody runs a setup like this, and even then the warning will last only for one release before we'll get a real error. block/raw-posix.c | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/block/raw-posix.c b/block/raw-posix.c index 844ac21..24d8582 100644 --- a/block/raw-posix.c +++ b/block/raw-posix.c @@ -503,6 +503,14 @@ static int raw_open_common(BlockDriverState *bs, QDict *options, error_setg_errno(errp, -ret, "Could not set AIO state"); goto fail; } + if (!s->use_aio && (bdrv_flags & BDRV_O_NATIVE_AIO)) { + error_printf("WARNING: aio=native was specified for '%s', but " + "it requires cache.direct=on, which was not " + "specified. Falling back to aio=threads.\n" + " This will become an error condition in " + "future QEMU versions.\n", + bs->filename); + } #endif s->has_discard = true; -- 1.8.3.1